laya-multilingual

Model overview

Non-autoregressive **System 1 decision model** covering 100+ languages. Give it a **state** (text, email, ticket, or JSON) and **typed questions**; it returns typed answers with probabilities in a single forward pass. No text generation, so nothing to parse and nothing to hallucinate. > **Long documents: `laya-multilingual` reads up to 8,192 tokens.** It ships with a 1,024-token limit that cuts long documents off, so pass `max_len=8192` for them: > > > > In the table below, 16 to 18 of 20 requests were answered correctly with up to about 4,000 tokens of text before them; beyond that results vary (8 to 17 of 20), so check long-document accuracy on your own data. Short inputs give identical answers with `max_len=8192`, and speed follows the input's real length, not the limit: short inputs are unchanged, and a 4,000-token input takes about 1.7 s on an Apple GPU. The default `Router()` keeps both `english` and this checkpoint resident, so a mixed workload no longer swaps checkpoints on every language change. For a server, load them up front so even the first request of each language is just a forward pass:

Key features

特点摘要待依据来源补齐，不凭模型名称推断能力。

Limitations & considerations

Sources are insufficient to establish specific limitations. Verify task performance, latency, data policies, licensing and version lifecycle.

Source prices & cost analysis

Prices vary by channel, region and configuration. Channel quotes are listed independently and are not treated as a single official provider price.

Public benchmarks & run configuration

暂无已采集的可核验公开评分。模型未上榜不代表能力较低。

Frequently asked questions

Sources & traceability

许可证：apache-2.0。模型介绍与第三方评测均以原来源的具体版本和方法为准。